Skip to content

TOOLS.md

Latest commit

 

History

History
60 lines (41 loc) · 2.52 KB

TOOLS.md

File metadata and controls

60 lines (41 loc) · 2.52 KB

TOOLS.md - Local Notes

Skills define how tools work. This file is for your specifics — the stuff that's unique to your setup.

What Goes Here

Things like:

  • Camera names and locations
  • SSH hosts and aliases
  • Preferred voices for TTS
  • Speaker/room names
  • Device nicknames
  • Anything environment-specific

Examples

### Cameras

- living-room → Main area, 180° wide angle
- front-door → Entrance, motion-triggered

### SSH

- home-server → 192.168.1.100, user: admin

### TTS

- Preferred voice: "Nova" (warm, slightly British)
- Default speaker: Kitchen HomePod

Why Separate?

Skills are shared. Your setup is yours. Keeping them apart means you can update skills without losing your notes, and share skills without leaking your infrastructure.

Add whatever helps you do your job. This is your cheat sheet.

Voice reply path

  • Edge TTS installed in venv: /home/mat/.openclaw/tools/edge-tts-venv
  • Edge TTS binary: /home/mat/.openclaw/tools/edge-tts-venv/bin/edge-tts
  • User-space ffmpeg binary: /home/mat/.openclaw/tools/edge-tts-venv/lib/python3.12/site-packages/imageio_ffmpeg/binaries/ffmpeg-linux-x86_64-v7.0.2
  • Tested working on 2026-03-09: text -> mp3 -> ogg/opus conversion succeeded
  • Important follow-up learned on 2026-03-09: openclaw agent --deliver can move media back to WhatsApp, but it still runs an agent turn rather than acting like a raw send primitive. So for voice-note delivery, verify not just media generation but the exact delivery behavior/content before treating the loop as finished.
  • Observed on 2026-03-11: longer WhatsApp voice replies can get truncated around ~1:30–1:50. Practical mitigation: keep spoken replies to roughly <= 900 chars / <= 90 sec and intentionally split longer answers into multiple voice notes.
  • Reinforced on 2026-03-13 after repeated truncation complaints: treat chunking as a hard default, not an occasional workaround. If a spoken answer might run long, proactively split it before sending.

YouTube transcript / video text path

  • Video text venv: /home/mat/.openclaw/tools/video-text-venv
  • Installed tools: youtube-transcript-api, yt-dlp
  • Fast transcript fetch script: /home/mat/.openclaw/workspace/scripts/youtube_transcript_fetch.py
  • Best current workflow on this machine:
    1. use youtube_transcript_fetch.py for direct transcript extraction when captions exist
    2. use yt-dlp as fallback for tougher YouTube retrieval cases
    3. if captions are missing, add audio-download + transcription as the next fallback instead of stopping